Loading ...
Sorry, an error occurred while loading the content.

Re: [linux] Clustering

Expand Messages
  • ed@ednevitible.co.uk
    On Fri, 30 Apr 2004 15:20:57 -0500 ... Well it all depends on loads. Sometimes throwing in new PCs is just throwing hardware at the problem. Why not just build
    Message 1 of 16 , May 1 8:09 AM
    • 0 Attachment
      On Fri, 30 Apr 2004 15:20:57 -0500
      "EXT-Reimer, Jim D" <jdr@...> wrote:

      > <snip>
      > > going to suffer, it would be more responsible - both towards
      > > the customers *and* the company itself - to refuse to carry
      > > out the project in question.
      >
      > That can be a good way to start looking for a new job too.

      Well it all depends on loads. Sometimes throwing in new PCs is just
      throwing hardware at the problem. Why not just build a cluster and add
      additional PCs when loads become high. Surely thats better than taking
      down a service for a few minutes. Replacing a P100 with a 2ghz model
      means taking down the service, if you could add another PC this month,
      and the following month then you have better reliability IMO. Speaking
      as someone who has not done this yet of course, but I spoke a fair bit
      about it at my last job, and then left to take a better position
      elsewhere, where I get to practice things rather than just talk.

      --
      icq: 21156382 msn: cryptic_edd at gmx dot net
    • ed@ednevitible.co.uk
      On Fri, 30 Apr 2004 23:19:38 GMT ... So far Ive built them quite a few mail servers. They re a hosting company and well, Im kinda replacing everything with
      Message 2 of 16 , May 1 8:11 AM
      • 0 Attachment
        On Fri, 30 Apr 2004 23:19:38 GMT
        "Wolf Dreamer" <miyaca@...> wrote:

        > They may need to reconcider their position. But then, Linux has a
        > rather unique definition of junk boxen. How bad are we talking about?
        > We may also
        > get lucky and they are going to use the lousy boxen for the prototype
        > then get it together for the real thing.

        So far Ive built them quite a few mail servers. They're a hosting
        company and well, Im kinda replacing everything with linux versions of
        their MS services. Although they're a MCSP (MS Solutions Certified
        Partner or something, dont quote me on the term). So as far as I can
        see, all its costing them is my salary, if the test versions of my boxes
        dont work out, then nothing has been lost other than a little bit of
        time taken to develop these things.


        --
        icq: 21156382 msn: cryptic_edd at gmx dot net
      • Wolf Dreamer
        H-availability clustering (as opposed to High Performance Computation) is not hard once you understand a few points. (1) you can not have the same IP on a
        Message 3 of 16 , May 9 8:47 AM
        • 0 Attachment
          H-availability clustering (as opposed to High Performance Computation) is
          not hard once you understand a few points.

          (1) you can not have the same IP on a network from two different hosts at
          the same time. It would confuse the ARP tables. Besides as soon as the
          second NIC card gets activated and an ARP Request is initiated, a response
          will tell it that another host has that IP and it will shutdown that
          interface.

          (2) You can two hosts for free using RedHat's Piranha or a program called
          Heartbeat. I use Heartbeat with a disk array for a high availability
          storage system. Three costs money - lots of money.

          (3) Heartbeat works with each host having its own IP and migrating the
          target IPs between them.

          HTH
        Your message has been successfully submitted and would be delivered to recipients shortly.